Conversion Rate Optimization (CRO)CRO foundations · Lesson 2 of 20
Funnel diagnosis: finding the biggest opportunities
Video lecture
Funnel diagnosis: finding the biggest opportunities
The narrated lecture is in production
Every chapter is scripted and ready. Browse the chapters and read the full transcript now — the video will appear here when it’s published.
Chapters
Transcript of the narration, chapter by chapter.
0:00 Funnel diagnosis
Imagine a bucket with holes in it. You could keep pouring in more water, more traffic, more ad spend. Or you could find the biggest hole and patch it. Funnel diagnosis is how you find the holes, and how you tell a big hole from a small one. In this lecture you'll learn to map a funnel, quantify where the value leaks, segment until the real problem appears, and check technical health before blaming design or copy. Then you'll watch me build a funnel exploration and rank leaks with a short script.
0:40 Why diagnose first
Why diagnose before you test? Because testing capacity is limited. Most sites can only run a handful of reliable tests each quarter. If you spend them on the wrong page, you've wasted months. Diagnosis tells you where the biggest opportunity is, which segments are struggling, and whether the problem is broken technology, missing information, or a weak offer. It also protects you from redesigning the homepage when the real leak is a payment method that fails on one browser.
1:14 Map the funnel
Start by mapping the funnel. For an online store: landing, product view, add to cart, checkout start, payment info, purchase. For lead generation: landing, form start, form submit, qualified lead, sale. For software: sign-up, activation, the aha moment, paid conversion. Write down the steps, and make sure each has a tracked event. Think of it like mapping a river before looking for leaks in the dam. You need to know where the water flows before you can see where it escapes.
1:49 Quantify the leaks
Now quantify. For each step, calculate the pass-through rate: how many people reach the next step. Then ask two questions. How many people drop here? And how feasible is an improvement? Here's a surprising fact. In a linear funnel, a ten percent relative improvement at any single step lifts final conversions by about ten percent. So the question isn't which step has the most drop-offs. It's where a ten percent improvement is most achievable. A step where only fifteen percent pass through usually has far more headroom than a step at seventy-two percent.
2:29 Segment until it appears
Next, segment until the problem appears. Averages hide everything. Split by device first: mobile and desktop often behave completely differently. Then by traffic source, new versus returning visitors, country, browser and landing page. A checkout that converts well on desktop but terribly on one mobile browser isn't a persuasion problem. It's a bug. A product page that converts well for search traffic but badly for social traffic might be a message-match or awareness problem. The segment tells you what kind of problem you have.
3:06 Technical health first
Before blaming design or copy, check technical health. Page speed, especially on mobile: look at Core Web Vitals, where Interaction to Next Paint replaced First Input Delay in twenty twenty-four. Errors in the console or in form validation. Payment methods failing for certain cards, wallets or countries. Broken layouts in right-to-left languages. And tracking accuracy. Fixing a broken payment method isn't glamorous, but it's often the single biggest win in a CRO programme.
3:38 Simple example: Pakistani electronics retailer
A simple example. An electronics retailer in Pakistan sees a big drop between checkout start and purchase. Segmenting by payment method reveals that cash on delivery works fine, but card payments fail far more often than expected on one mobile browser, because a 3D Secure pop-up is blocked. That's not a design problem. It's a bug with a clear fix. After the payment provider's recommended integration is applied, completion for card payments recovers. No test needed. Just diagnosis and a fix.
4:13 Realistic example: UK fashion brand (illustrative)
Now a realistic scenario with illustrative numbers. A UK fashion brand gets forty thousand mobile visits from paid social in a month. Twenty-six thousand view a product, three thousand nine hundred add to cart, two thousand three hundred start checkout, one thousand five hundred enter payment info, and one thousand and eighty purchase. The product view to add-to-cart step passes only about fifteen percent. That's the headroom. Recordings filtered to product page exits show people tapping the size selector repeatedly and scrolling to find delivery information. Survey answers mention being unsure about fit. So the diagnosis is clear: product pages on mobile lack fit guidance and delivery clarity. That becomes the first research-backed hypothesis.
5:03 Watch me do it: GA4 funnel + leak script
Watch me do it. In GA4, I open Explore and choose a funnel exploration. I add steps using recommended events: session start, view item, add to cart, begin checkout, add payment info, and purchase. I break it down by device category, then by default channel group. I export the step counts and paste them into a short Python script from the lesson text. It prints each step's pass-through rate, the drop-off, and what a ten percent improvement at that step would add in orders and revenue. Then I sort by feasibility, using research, not just numbers. The script tells me the size of the prize. Research tells me whether I can win it.
5:52 From 'where' to 'why'
Numbers tell you where people leave. They rarely tell you why. So every funnel diagnosis should end with a hand-off to qualitative research, targeted at the leak you found. If the leak is on mobile product pages from social traffic, filter session recordings to exactly that segment, and watch twenty to thirty sessions with a question in mind. Add a one-question on-page survey on those pages: what's stopping you from adding this to your bag today? Look at support tickets and chat logs that mention product details. By the time you write a hypothesis, you want the what from analytics and the why from people, pointing at the same spot. That's called triangulation, and it's the difference between a guess and a finding.
6:45 AI in diagnosis
AI can speed this up. Once you've exported funnel tables with no personal data, an assistant can spot segment differences, draft charts and summarise findings. Some analytics tools include built-in AI insights too. Two rules. Ask the assistant to show the calculation behind every number, and check a few by hand, because models can misread tables. And treat AI insights as prompts to investigate, not conclusions. The model might notice that Android users convert less. It can't tell you it's because your date picker is broken on one keyboard. Recordings can.
7:25 Common mistakes
Common mistakes. Looking only at site-wide averages. Chasing the step with the most drop-offs instead of the most achievable improvement. Ignoring technical issues and jumping to redesigns. Using an open funnel when you meant a strict one, so the numbers don't mean what you think. Comparing different time periods without accounting for seasonality or campaigns. And treating analytics as perfectly accurate when consent and tracking gaps exist.
7:54 Recap
Recap. Map every step of the funnel and make sure each is tracked. Quantify pass-through rates, remembering that the question is where improvement is most achievable. Segment by device, source and browser until the problem type appears. Check technical health first, because fixing what's broken is often the biggest win. Use AI to speed up analysis, with verification. Try this now: build a strict funnel exploration for your main conversion path, break it down by device, run the leak script from the lesson text, and write down the one step you'd research first, and why.
Start where the money leaks
Most sites have dozens of pages and hundreds of possible improvements. CRO teams that start by redesigning the homepage often waste months. Instead, map the funnel, quantify each leak, and focus on the steps where improvement is both likely and valuable.
Map the funnel
List the steps from first visit to value, with the event that marks each step:
E-commerce: Landing -> Product page -> Add to cart -> Checkout start -> Shipping -> Payment -> Purchase
Lead gen: Landing -> Service page -> Pricing/Case study -> Form start -> Form submit -> Qualified lead
SaaS: Landing -> Pricing -> Sign-up start -> Account created -> Activated (key action) -> PaidFor each step, record volume and step conversion rate, segmented by device and main traffic sources. Aggregate funnels hide problems: a checkout that converts well on desktop but poorly on mobile looks "average" overall.
Quantify the opportunity
A simple way to compare leaks is to estimate the value of improving each step:
Opportunity = Traffic at step x Realistic improvement x Downstream conversion x Value per conversionIllustrative example for a lead-generation site (numbers invented for teaching):
| Step | Monthly users | Step rate | Downstream to qualified lead | Idea |
|---|---|---|---|---|
| Service page -> form start | 8,000 | 10% | 30% | Clarify offer and proof |
| Form start -> submit | 800 | 45% | 67% | Reduce fields, fix mobile keyboard |
| Submit -> qualified | 360 | 50% | — | Qualification questions |
Improving form completion from 45% to 55% adds 80 submissions and roughly 40 qualified leads. Improving the service page from 10% to 11% adds 80 form starts, about 44 submissions and roughly 22 qualified leads. The form fix looks more valuable and easier — a strong first candidate.
Page-level prioritisation
Beyond funnel steps, rank templates and pages by:
- Traffic — more visitors means faster tests and bigger impact.
- Value — pages close to money (pricing, product, checkout) matter more.
- Gap — how much worse the page performs than comparable pages or segments.
- Ease — templates (all product pages) let one change affect thousands of URLs.
A product page template on a large store is often the highest-leverage place to work because one change applies everywhere.
Segment until the problem appears
Useful segments for diagnosis:
| Segment | Why it matters |
|---|---|
| Device (mobile / desktop / tablet) | Layout, speed and input differences |
| Traffic source | Intent differs: brand search visitors behave differently from cold social traffic |
| New vs returning | First-time visitors need more trust and explanation |
| Country / language | Currency, delivery, payment methods, right-to-left layouts |
| Landing page | Message match with ads or search queries |
| Browser | Rendering bugs |
When one segment's conversion rate is dramatically lower than others with no obvious reason, you have found a research lead.
Technical health first
Before sophisticated research, rule out technical problems:
- Page speed and Core Web Vitals on real mobile devices and networks common in your markets.
- Browser and device bugs — test key journeys on popular devices, including mid-range Android phones.
- Form validation errors — track validation failures as events if possible.
- Payment failures — check gateway decline reasons; missing local payment methods (for example cash on delivery, local wallets or buy-now-pay-later options where they are popular) can be a major leak.
Worked example: a Pakistani electronics retailer
Funnel analysis shows mobile users add to cart at similar rates to desktop but complete checkout far less often. Segmenting by payment step reveals a large drop at payment for mobile users. Session recordings (next module) show the card form's expiry field opens a full keyboard and rejects common formats. Fixing the input type and accepting multiple formats is implemented immediately — no test required — and mobile checkout completion is monitored for improvement.
Hands-on: a funnel exploration in GA4 and a leak calculator
In GA4, use Explore → Funnel exploration. A typical ecommerce funnel uses recommended events:
Step 1 session_start (or page_view on a landing page)
Step 2 view_item
Step 3 add_to_cart
Step 4 begin_checkout
Step 5 add_payment_info
Step 6 purchase
Breakdown: device category; then session default channel group
Toggle: "Make open funnel" off (users must start at step 1) for a strict funnelExport the step counts, then estimate where improvement is worth most. This short script ranks leaks by the revenue a realistic improvement would add (illustrative inputs):
steps = [ # (step name, users reaching step) - last 28 days, mobile, paid social
("Landing", 40_000), ("Product view", 26_000), ("Add to cart", 3_900),
("Checkout start", 2_300), ("Payment info", 1_500), ("Purchase", 1_080),
]
aov = 62.0 # average order value
relative_gain = 0.10 # "what if this step's pass-through improved by 10%?"
final = steps[-1][1]
print(f"Baseline purchases: {final:,} revenue: {final * aov:,.0f}")
for (name_a, a), (name_b, b) in zip(steps, steps[1:]):
rate = b / a
extra = final * relative_gain # a 10% lift at any single step lifts purchases ~10%
print(f"{name_a:>14} -> {name_b:<14} pass-through {rate:6.1%} "
f"drop-off {a - b:>7,} +10% here = +{extra:,.0f} orders (~{extra * aov:,.0f})")Notice that a 10% relative improvement at any step adds roughly the same number of orders. What differs is how feasible a 10% improvement is: a step with 15% pass-through (product view → add to cart above) usually has far more headroom than one with 72%. Combine the numbers with research, not instead of it.
Using AI to speed up diagnosis
Once you have exported funnel tables (no personal data), an AI assistant can help you spot segment differences, draft charts and summarise findings. Ask it to show the calculation for every number it reports, and check a few by hand — models can misread tables. Some analytics tools also offer built-in AI insights; treat them as prompts for investigation.
Common mistakes
- Looking only at site-wide conversion rate.
- Starting with the homepage because it is the most visible page internally.
- Ignoring technical issues and jumping straight to persuasion.
- Estimating opportunity with unrealistic improvement assumptions.
Diagnosis checklist
Key takeaways
- Map the funnel and quantify each leak before choosing what to work on.
- Segment by device, source and visitor type — averages hide problems.
- Templates and pages close to money usually offer the highest leverage.
- Rule out technical issues before persuasion work.
Check your understanding
Quick questions to lock in the lesson. They don’t count towards your certificate.
Put it into practice
Map your main funnel with step rates by device, estimate the opportunity for the top two leaks using the formula, and pick one to research first.
Enrol for free to save your progress
Reading is always free. Enrol to keep your place, take the final assessment and earn a verifiable certificate.